Connected data
We retain originals collected through files, APIs and browsers, then organize and validate them against shared standards. We manage how products, orders, inventory and settlements connect.
TECHNOLOGY / SYSTEMS & PROCESS
Connecting data and work:
the foundation for AX.
Data collection and validation, workflow automation and traceable decision support.
Explore the software foundation connecting daily operations, and our content creation technology.
AX · From data to work
For AI to be useful,
work needs connected context.
We connect data across tasks and automate repetitive processing.
Summaries and suggestions are designed to retain evidence for people to review.
Connected execution
We assign jobs around schedules and resources, and track status, failures and retries. Our own operating foundation connects each stage of the work.
Traceable decisions
We build paths from metrics, reports and AI suggestions back to original evidence. We validate scope for each workflow and record human reviews and decisions.
CDH operational data and ioCAST production data are handled within each service’s responsibilities and permissions. The production diagram below does not describe all of CDH.
CONTENT PRODUCTION / SYSTEM ARCHITECTURE
Separate responsibilities.
Connected sources and outputs.
Storage handles files. Engines handle production.
ioCAST handles review and selection.
01 / INPUT
Source files
Recorded video · Photos · Audio
Production assets as files
02 / STORE
Dropfile
Files · Versions · Checksums
Pass source references
03 / MAKE
Production tools
EngineScript → Cuts & captions → Render
AI musicStyle & lyrics → Audio candidates
04 / REVIEW
ioCAST
Review frames, captions and evidence
Compare candidates → Human selection
05 / OUTPUT
Selected output
Video · Music · Thumbnails
Export · Storage references
↳ Selected outputs are stored in Dropfile and linked back to the sources used in production.
Verified product information · Brand materialsEvidence for Engine production and ioCAST review
02 / VIDEO PIPELINE
One video.
Six production stages.
From input to final frame,
we connect the information each stage records.
- 01Assets · Products · Profiles
Fix inputs and production criteria
We link recorded assets, product descriptions, brand voice, logos and profile versions to the production request. We define the materials and purpose first.
- 02Scene information · Evidence
Analyze scenes and find evidence
We analyze recorded scenes and find product and brand materials for the script. Source references and verified information remain linked as production evidence.
- 03Copy sets · Caption candidates
Build script and caption candidates
We compose text from product information and assets. We are improving review of the sequence and role of the whole piece, alongside individual claims.
- 04Timeline · Audio
Arrange cuts, captions and audio
We combine scene order and duration, caption placement and music segments. Music is selected through a separate generation and listening workflow, then linked to the video.
- 05Video · Frames · Thumbnails
Render and inspect actual frames
We render with Remotion- and FFmpeg-based tools. We inspect final captioned frames, crops and zooms, and prepare outputs such as thumbnails.
- 06Selected output · Production history
Compare, select and retain history
People review candidates in ioCAST and select an output. Production runs, source references and output files are linked for later retrieval.
Clip analysis · In development
We are improving clip analysis and scene selection. We reuse shared analysis of the same file while keeping product-specific assessments separate, and are preparing a library for reusing crop, zoom and speed variants.
03 / JUDGMENT & EVIDENCE
What does an assessment check,
and what does it record?
- 01Facts · Figures · Sources
Match copy to product information
We check claims against verified product information and figures. We look for unsupported statements about materials, performance or benefits that scenes alone cannot establish.
- 02Context · Sequence · Repetition
Overall flow and each sentence’s role
We review the order and role of the complete caption set, checking for contradictions, unnecessary repetition and wording that conflicts with the product context.
- 03Scenes · Crops · Captions
Match wording to the actual render
We review captions alongside actual scenes. We inspect the rendered output to check whether crops, zooms and caption placement suit the selected wording.
- 04Warnings · Evidence · Revisions & selection
Warnings and human judgment
We have defined warning categories for exaggerated or prohibited wording, implied medical benefits and unsupported figures. We evaluate missed risks alongside unnecessary warnings.
Final checks for advertising use
People verify evidence for claims, required advertising disclosures and channel and industry standards. An absence of warnings does not guarantee approval or factual accuracy.
04 / COLLECTION & FILE PROCESSING
Files and APIs
enter through different paths.
Files connect through storage references.
API and web data connect directly to processors.
A / Direct connection
Collected data without a file upload
DIRECT INPUT
API · Web · Blogs
Authenticated API responses
Collect text, media and sources
PROCESS
Data processing
Organize fields · Parse · Validate
Retain source and original information
USE
Product information
Lookups · Production evidence
Connect verified information to production
B / File-based
Three routes from shared storage, depending on purpose
Select a route to replay its flow.
SOURCE / MEDIA
Video · Image · Audio
Source assets for production
SOURCE / DATA
Excel · CSV · Images
Product data · OCR images
SOURCE / FILE
CSV · TXT · Excel
Files for encoding conversion
STORE & REFERENCE
Dropfile
File storage · Versions · Checksums
Pass file references without interpreting content
MAKE
Engine · Music production
Scene analysis · Editing · Generation
READ
Data processing · OCR
Parse · Extract text · Validate
CONVERT
File converter
Detect encoding · Convert formats
OUTPUT & REUSE
Reusable outputs
Production candidatesioCAST review → Store selected outputs
Product information to verifyCheck against originals → Lookups and production materials
Converted filesDownload → Clean up temporary storage
Video · Image · Audio
Source assets for production
- STOREDropfile
File storage · Pass references
- PROCESSEngine · Music production
- OUTPUTProduction candidates
ioCAST review · Store selected outputs
Excel · CSV · Images
Product data · OCR images
- STOREDropfile
File storage · Pass references
- PROCESSData processing · OCR
- OUTPUTProduct information to verify
Check originals · Lookups and production evidence
CSV · TXT · Excel
Files for encoding conversion
- STOREDropfile
File storage · Pass references
- PROCESSFile converter
- OUTPUTConverted files
Download · Clean up temporary storage
↳ Dropfile stores originals; processors interpret them. Production outputs and converted files link back to storage.
Storage and interpretation boundaries
Dropfile stores files; dedicated processors handle parsing, OCR and conversion. Photo OCR results are checked with originals and sources, and people review suggested categories and fields. API integrations align service authentication and received fields.
Current product input scope
Collection, OCR and API integration are existing capabilities. ioCAST product registration focuses on file uploads and text entry; direct import from product URLs is not currently available. CSV and TXT encoding conversion and CSV extraction from Excel’s first data sheet use a separate file processing workflow.
05 / STORAGE FOUNDATION
Dropfile.
Through every file state.
We store source video, images, music, finished outputs and thumbnails,
and manage the file lifecycle from upload to restoration.
Multipart uploads · Retries
We split large uploads and track progress. Retained upload parts support continued retries.
Versions · Checksums
We manage file sizes, versions and checksums to link production sources and outputs to the exact files used.
Previews · Thumbnails
We provide previews for supported formats and thumbnails for images and video. Thumbnail processing status and failures are tracked too.
Trash · Version restoration
Deleted files and previous versions can be restored according to file state and retention policy.
Batch actions · Downloads
People can select multiple files for processing and prepare batch downloads. Long jobs such as compression expose their status.
Authenticated storage API
APIs connect uploads, lookups and downloads. Access controls distinguish file owners and administrators.
06 / EXECUTION & ACCESS
The scheduler coordinates
job order and resources.
REQUEST
Request & queue
Request video and voice jobs
Add to queue
SCHEDULE
Priorities · Resources
Adjust job order
Assign available resources
EXECUTE
Execution · Status tracking
Execution, interruption and completion history
Verify uncertain termination
RECORD
Results · History
Store output files
Link requests and execution records
Verify execution before moving on
We coordinate GPU resources and job order for video and voice work. The design holds subsequent execution until a job with uncertain termination has been checked.
Organization scope and request permissions
We check permissions for sessions and API requests, separating data access by organization. Shared account and tool integrations are being prepared to preserve permissions and execution history.
Asset usage terms and records
We manage filmed participants, contract materials and usage periods, and record assets and contract information used in each request. File versions and production runs are tracked together.
This does not mean every product runs automatically in production. Integration and activation are verified separately for each consuming service.
VOICE & MUSIC
Voice, sentence by sentence.
Music, by mood and segment.
We design the two production workflows separately
and connect them on the video timeline.
Narration
- Text & pronunciation
We distinguish on-screen text from spoken text. A dictionary manages pronunciation for proper names and units.
- Sentence generation and reuse
We record voices and generation settings, storing audio by sentence. We are connecting a workflow to regenerate only the sentences needed and reuse outputs.
- Synthesis and quality validation
We have tested voices, delivery styles and sentence joins. Voice consistency in long scripts, pronunciation, silence and duration require further validation.
Full narration production is not generally available yet. We are validating scheduler and ioCAST integration as we apply it.
Music generation and LoRA
- Style · Lyrics · Vocal direction
Mood, style, lyrics and vocal settings guide the generation of audio candidates.
- Optional LoRA
Compatible music generation routes let people select a LoRA and its strength. They can listen to and compare the result with base generation.
- Segment selection and usage history
People use waveforms to inspect segments and select music for video. Usage counts and recent usage history are available.
Music LoRA is separate from training or tuning assessment technology. It does not imply better quality across all styles.
07 / DEVELOPMENT & NEXT
What we are refining now.
What we will connect next.
Development stage and availability are stated separately.
Preview work in progress
We have developed an interface for listening to selected music and distinguishing edits in progress from the original output.
Merge verified on 2026-10-23 · Production deployment requires separate verificationTechnology adapted to the task
We compare results on fixed materials and review quality and resource use. Content, format, processing time and failures are considered together; actual frames and listening results determine where to apply the technology.
MCP connects tools and data
We are preparing connections for conversational AI to query permitted data and call production tools. Starting with read-focused validation, we plan to expand to authorized production requests while preserving user and organization permissions and execution history. Production-specific tuning and proprietary technology research are also planned.
Beyond production
We aim to extend into long-form production, channel and advertising integration, and feeding results into future planning. Our own information and partner-content channels are being prepared as internal operating projects.
Carry brand standards forward
ioCAST brand profiles manage the identity and standards used in production. Broader brand asset management is being prepared as a separate plan.
Technology for understanding products
We provide reference information for reviewing product size, fit and measurement differences between sizes. Product interface integration is in progress; final checks use physical samples and fittings.
Based on documents reviewed on 2026-10-24. Availability will be communicated after feature-specific integration and validation.
The next step in how you work
Which task
takes the longest?
Let's find the work you can reduce in your current workflow.